Missing adapter config file prevents GGUF conversion
Hi, this project seems very cool and interesting.
I am trying to convert the Trace-Inverter-4B model to GGUF format using the convert_hf_to_gguf.py script from llama.cpp.
The conversion script fails with a ValueError regarding tensor mapping:
ValueError: Can not map tensor 'model.layers.0.mlp.down_proj.base_layer.weight'
I believe this occurs}s because the repository contains the unmerged safetensors weights with the PEFT wrapper keys (like .base_layer.weight, .lora_A, and .lora_B) instead of the standard hf layout.
To manually merge these low rank adaptation matrices and standardize the tensor keys, I would need the alpha and rank values to calculate the scaling factor.
However, the adapter_config.json file is not present in the repository files.
Could you please also upload the adapter_config.json file used during the making of the lora, or share the lora_alpha and r parameters used during training?
Thank you.